Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·26d ago·5 min read
Pseudo-Label Augmentation for Affect Sensing in Small Collaborative Groups
TL;DR
Pseudo-label augmentation enhances affect sensing in small collaborative groups by effectively utilizing sparse physiological data. The approach outperforms traditional methods, particularly in known-team settings.
✦ Why It Matters
Implement pseudo-label augmentation techniques to improve affect sensing in your collaborative projects today.
Key Takeaways
- Utilize pseudo-label augmentation to enhance affect sensing accuracy.
- Leverage the GroupAffect-4 dataset for collaborative group studies.
- Apply personality similarity as a team membership filter.
- Explore cross-subject transfer methods for better model generalization.
Full Summary
Affect sensing in group interactions often suffers from limited labeled data, as wearable devices generate extensive physiological data while self-reports are infrequent. This study utilizes the GroupAffect-4 dataset, which includes physiological data, eye tracking, personality traits, and affect labels, to explore pseudo-label augmentation techniques.
Various methods were compared, including Gaussian Process pseudo-labeling and personality-aware trust weighting. Results indicated that pseudo-label augmentation significantly improved performance over a baseline that used only labeled data, especially in known-team scenarios.
However, personality similarity was found to be more effective as a filter for team membership rather than a trust signal. The study also noted that while cross-subject transfer showed promise, strict session isolation diminished the benefits of augmentation.
Overall, the findings suggest that pseudo-label augmentation can optimize the use of limited collaborative affect data, with personality information serving as a useful selection tool within teams.